Quo Vadis#

That’s actually the wrong person. Quo vadiums, where are we going, would be more accurate. My previous post was about all the energy laying ready to release for an economic event. I tried to hint the dip, recession, slow-down, stagflation, or depression might take years to play out. In fact, we may be in it now. Or this may be a prelude. Or it may have started three months ago. It’s hard to know until well after the fact.

What happened after the dot-com bust? A lot of firms paid a lot of money to lay a lot of fiber prior to the bust. Suddenly, there was a lot of fiber laying dark. Getting internet connectivity for your business was no longer a seller’s market. It was a buyers market, making some new ideas feasible. YouTube could cheaply light up dark fiber. Bandwidth, data center space, and office space suddenly wasn’t the issue it was just a year or two earlier.

Some folks left technology. I went to more sales and back to school, in a way. But it also released resources. People who had been scrambling to meet deadlines for dead end projects, like Pets.com, were suddenly free. They re-mixed and out came the Web 2.0 companies. For good or ill, the collapse of the first internet boom made space for the second internet boom. Combined with lower interest rates, it created space.

What will be left with, other than data-center sized cement pads all over the United States? I don’t know what we can do with the accelerator chips. If the market loses the belief that untold wealth will come from AI, a 30,000 USD chip has limited appeal. It may be that AI services around smaller models that train faster might still be attractive. But the insane projecttions around 100s of gigawats of capacity will fall away. That will leave a lot of chips sitting in someone’s inventory. And chips age like milk. After a couple of years, the new chips have better power performance. Older chips get heavily discounted.

In addition to the AI chips, which have limited utility outside of AI, there will also be general purpose servers. AI requires regular computer and storage as well as accelerators. The market for used parts coming out of data centers may get flooded after a while. I think there is a business in here, somewhere. But a lot of the servers themselves are built to specs few businesses will run. They’re not simple HP rack server, coming from companies that build semi-custom designs for hyper-scalers.

Depending on how long this takes, we might be slightly over-invested in power generation. It takes time to build out power generation and transmission. While there is a mix of sources, data centers seem to prefer gas-turbine generators. It may not be the case we have a glut of renewables, allowing companies to decomission gas plants. There may be limited demand for a gas-turbine generator next to a big cement pad.

I don’t know if quantum computing will be a beneficiary. But it takes a lot of energy to keep something absolute zero. It also takes space. having a bunch of empty data centers, with a lot of power capcity, and a lot of thermal dissipation capacity, might be attractive. I don’t really know if there’s a future in quantum. Whether or not there is a sufficiently large market for the kinds of problems quantum computers solve, I don’t know.

Depending on the degree of the economic downturn, cloud may become too expensive for some companies. They may decide to run their own computers. Especially if ther there are plenty of admins and developers looking for jobs. Or some the tasks are automated using the small, focused AI models they can run on local hardware. What I think is stopping most companies is the complexity of running infrastructure. I don’t think a lot of that complexity is necessary. If the tooling improves, the economy of running your own hardware may be better.

The counter trend is the growth in ARM usage. ARM chips tend to be tailored, unlike x86 chips. An ARM chip made for Microsoft Azure isn’t necessarily the same as an ARM chip made for AWS. In addition, there may be additional priprietary chips on that motherboard. It could well be that there isn’t much, or any, market for used ARM processors. The AWS ARM processor may not really work outside of AWS.

One possible use for the cement pads are private prisons. I have to think an empty data center with power, cooling, and water is fine for a private prison. Why would we need the huge increase for private prisons, away from population centers? To hold the deportees the administration is rounding up.

I’m more hopeful. Football field sized cement pads. Ample power. Ample Networking. Maybe we could put computer controlled manufacturing equipment in them. Hope on your CAD program and draw a part. A couple of days later, it shows up at your door. Laser cut, milled, water-jet cut, CNC machined, etc. But who knows? That would mean we actually start making things again. Insted relying on number go up infinite money glitches.